Use this URL to cite or link to this record in EThOS:
Title: Design and performance characterisation of a modular surveillance system for a distributed processing platform
Author: Robinson, Mike
Awarding Body: University of Edinburgh
Current Institution: University of Edinburgh
Date of Award: 2001
Availability of Full Text:
Access from EThOS:
Full text unavailable from EThOS. Please try the link below.
Access from Institution:
This thesis investigates pedestrian monitoring using image processing for state-of-the-art real-time distributed camera-processor architectures. An integrated design and evaluation process is proposed, where the surveillance task is analysed into component modules, each corresponding to a self-contained vision process. Different approaches to each process are implemented independently, using Object-oriented design principles to facilitate both system construction and module interchange during comparative testing. Standard algorithms, from the computer vision literature, together with novel variants are used but the scope is restricted to what can be implemented to run in real-time, on a modest image processing engine. Comparison is made between median-based and mixture-of-Gaussian based methods for background representation, between connected component and boundary following object segmentation and between pixel-based and 2-D model based (PCA with cubic splines) methods for object classification. Quantitative performance-characterization data for existing solutions is not generally available, in the literature, in the form of bench-mark test-sequences and is time-consuming and costly to produce, for novel methods. A substantial test-data-set of real surveillance image-sequences has been acquired, to test; the system and compare alternative approaches. A novel performance-characterization technique is proposed: it offers comparative quantitative evaluation of the performance and resource requirements of a system. This approach is applied to the different system variants, comprised of the alternative module combinations. The results of running the system variants on the test data are compared against manually derived ground-truth data for pedestrian detection. The performance characterization approach provided clear comparative data on performance and resource requirements for each variant, analysed by scene and event type. From a review of these results, the optimum module combination is chosen: this is a system composed of median-based background representation, boundary following object segmentation and model-based object classification.
Supervisor: Not available Sponsor: Not available
Qualification Name: Thesis (Ph.D.) Qualification Level: Doctoral
EThOS ID:  DOI: Not available